i965: Let brw_flag_reg() choose the flag reg and subreg.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_eu.c
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
5
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
13
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keith@tungstengraphics.com>
30 */
31
32
33 #include "brw_context.h"
34 #include "brw_defines.h"
35 #include "brw_eu.h"
36
37 #include "glsl/ralloc.h"
38
39 /* Returns the corresponding conditional mod for swapping src0 and
40 * src1 in e.g. CMP.
41 */
42 uint32_t
43 brw_swap_cmod(uint32_t cmod)
44 {
45 switch (cmod) {
46 case BRW_CONDITIONAL_Z:
47 case BRW_CONDITIONAL_NZ:
48 return cmod;
49 case BRW_CONDITIONAL_G:
50 return BRW_CONDITIONAL_L;
51 case BRW_CONDITIONAL_GE:
52 return BRW_CONDITIONAL_LE;
53 case BRW_CONDITIONAL_L:
54 return BRW_CONDITIONAL_G;
55 case BRW_CONDITIONAL_LE:
56 return BRW_CONDITIONAL_GE;
57 default:
58 return ~0;
59 }
60 }
61
62
63 /* How does predicate control work when execution_size != 8? Do I
64 * need to test/set for 0xffff when execution_size is 16?
65 */
66 void brw_set_predicate_control_flag_value( struct brw_compile *p, GLuint value )
67 {
68 p->current->header.predicate_control = BRW_PREDICATE_NONE;
69
70 if (value != 0xff) {
71 if (value != p->flag_value) {
72 brw_push_insn_state(p);
73 brw_MOV(p, brw_flag_reg(0, 0), brw_imm_uw(value));
74 p->flag_value = value;
75 brw_pop_insn_state(p);
76 }
77
78 p->current->header.predicate_control = BRW_PREDICATE_NORMAL;
79 }
80 }
81
82 void brw_set_predicate_control( struct brw_compile *p, GLuint pc )
83 {
84 p->current->header.predicate_control = pc;
85 }
86
87 void brw_set_predicate_inverse(struct brw_compile *p, bool predicate_inverse)
88 {
89 p->current->header.predicate_inverse = predicate_inverse;
90 }
91
92 void brw_set_conditionalmod( struct brw_compile *p, GLuint conditional )
93 {
94 p->current->header.destreg__conditionalmod = conditional;
95 }
96
97 void brw_set_access_mode( struct brw_compile *p, GLuint access_mode )
98 {
99 p->current->header.access_mode = access_mode;
100 }
101
102 void
103 brw_set_compression_control(struct brw_compile *p,
104 enum brw_compression compression_control)
105 {
106 p->compressed = (compression_control == BRW_COMPRESSION_COMPRESSED);
107
108 if (p->brw->intel.gen >= 6) {
109 /* Since we don't use the 32-wide support in gen6, we translate
110 * the pre-gen6 compression control here.
111 */
112 switch (compression_control) {
113 case BRW_COMPRESSION_NONE:
114 /* This is the "use the first set of bits of dmask/vmask/arf
115 * according to execsize" option.
116 */
117 p->current->header.compression_control = GEN6_COMPRESSION_1Q;
118 break;
119 case BRW_COMPRESSION_2NDHALF:
120 /* For 8-wide, this is "use the second set of 8 bits." */
121 p->current->header.compression_control = GEN6_COMPRESSION_2Q;
122 break;
123 case BRW_COMPRESSION_COMPRESSED:
124 /* For 16-wide instruction compression, use the first set of 16 bits
125 * since we don't do 32-wide dispatch.
126 */
127 p->current->header.compression_control = GEN6_COMPRESSION_1H;
128 break;
129 default:
130 assert(!"not reached");
131 p->current->header.compression_control = GEN6_COMPRESSION_1H;
132 break;
133 }
134 } else {
135 p->current->header.compression_control = compression_control;
136 }
137 }
138
139 void brw_set_mask_control( struct brw_compile *p, GLuint value )
140 {
141 p->current->header.mask_control = value;
142 }
143
144 void brw_set_saturate( struct brw_compile *p, bool enable )
145 {
146 p->current->header.saturate = enable;
147 }
148
149 void brw_set_acc_write_control(struct brw_compile *p, GLuint value)
150 {
151 if (p->brw->intel.gen >= 6)
152 p->current->header.acc_wr_control = value;
153 }
154
155 void brw_push_insn_state( struct brw_compile *p )
156 {
157 assert(p->current != &p->stack[BRW_EU_MAX_INSN_STACK-1]);
158 memcpy(p->current+1, p->current, sizeof(struct brw_instruction));
159 p->compressed_stack[p->current - p->stack] = p->compressed;
160 p->current++;
161 }
162
163 void brw_pop_insn_state( struct brw_compile *p )
164 {
165 assert(p->current != p->stack);
166 p->current--;
167 p->compressed = p->compressed_stack[p->current - p->stack];
168 }
169
170
171 /***********************************************************************
172 */
173 void
174 brw_init_compile(struct brw_context *brw, struct brw_compile *p, void *mem_ctx)
175 {
176 memset(p, 0, sizeof(*p));
177
178 p->brw = brw;
179 /*
180 * Set the initial instruction store array size to 1024, if found that
181 * isn't enough, then it will double the store size at brw_next_insn()
182 * until out of memory.
183 */
184 p->store_size = 1024;
185 p->store = rzalloc_array(mem_ctx, struct brw_instruction, p->store_size);
186 p->nr_insn = 0;
187 p->current = p->stack;
188 p->compressed = false;
189 memset(p->current, 0, sizeof(p->current[0]));
190
191 p->mem_ctx = mem_ctx;
192
193 /* Some defaults?
194 */
195 brw_set_mask_control(p, BRW_MASK_ENABLE); /* what does this do? */
196 brw_set_saturate(p, 0);
197 brw_set_compression_control(p, BRW_COMPRESSION_NONE);
198 brw_set_predicate_control_flag_value(p, 0xff);
199
200 /* Set up control flow stack */
201 p->if_stack_depth = 0;
202 p->if_stack_array_size = 16;
203 p->if_stack = rzalloc_array(mem_ctx, int, p->if_stack_array_size);
204
205 p->loop_stack_depth = 0;
206 p->loop_stack_array_size = 16;
207 p->loop_stack = rzalloc_array(mem_ctx, int, p->loop_stack_array_size);
208 p->if_depth_in_loop = rzalloc_array(mem_ctx, int, p->loop_stack_array_size);
209
210 brw_init_compaction_tables(&brw->intel);
211 }
212
213
214 const GLuint *brw_get_program( struct brw_compile *p,
215 GLuint *sz )
216 {
217 brw_compact_instructions(p);
218
219 *sz = p->next_insn_offset;
220 return (const GLuint *)p->store;
221 }
222
223 void
224 brw_dump_compile(struct brw_compile *p, FILE *out, int start, int end)
225 {
226 struct brw_context *brw = p->brw;
227 struct intel_context *intel = &brw->intel;
228 void *store = p->store;
229 bool dump_hex = false;
230
231 for (int offset = start; offset < end;) {
232 struct brw_instruction *insn = store + offset;
233 struct brw_instruction uncompacted;
234 printf("0x%08x: ", offset);
235
236 if (insn->header.cmpt_control) {
237 struct brw_compact_instruction *compacted = (void *)insn;
238 if (dump_hex) {
239 printf("0x%08x 0x%08x ",
240 ((uint32_t *)insn)[1],
241 ((uint32_t *)insn)[0]);
242 }
243
244 brw_uncompact_instruction(intel, &uncompacted, compacted);
245 insn = &uncompacted;
246 offset += 8;
247 } else {
248 if (dump_hex) {
249 printf("0x%08x 0x%08x 0x%08x 0x%08x ",
250 ((uint32_t *)insn)[3],
251 ((uint32_t *)insn)[2],
252 ((uint32_t *)insn)[1],
253 ((uint32_t *)insn)[0]);
254 }
255 offset += 16;
256 }
257
258 brw_disasm(stdout, insn, p->brw->intel.gen);
259 }
260 }